fs/btrfs/file.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/mpage.h>
  27 #include <linux/falloc.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/slab.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "ioctl.h"
  38 #include "print-tree.h"
  39 #include "tree-log.h"
  40 #include "locking.h"
  41 #include "compat.h"
  42
  43
  44 /* simple helper to fault in pages and copy.  This should go away
  45  * and be replaced with calls into generic code.
  46  */
  47 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
  48                                          int write_bytes,
  49                                          struct page **prepared_pages,
  50                                          struct iov_iter *i)
  51 {
  52         size_t copied = 0;
  53         int pg = 0;
  54         int offset = pos & (PAGE_CACHE_SIZE - 1);
  55         int total_copied = 0;
  56
  57         while (write_bytes > 0) {
  58                 size_t count = min_t(size_t,
  59                                      PAGE_CACHE_SIZE - offset, write_bytes);
  60                 struct page *page = prepared_pages[pg];
  61                 /*
  62                  * Copy data from userspace to the current page
  63                  *
  64                  * Disable pagefault to avoid recursive lock since
  65                  * the pages are already locked
  66                  */
  67                 pagefault_disable();
  68                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
  69                 pagefault_enable();
  70
  71                 /* Flush processor's dcache for this page */
  72                 flush_dcache_page(page);
  73                 iov_iter_advance(i, copied);
  74                 write_bytes -= copied;
  75                 total_copied += copied;
  76
  77                 /* Return to btrfs_file_aio_write to fault page */
  78                 if (unlikely(copied == 0)) {
  79                         break;
  80                 }
  81
  82                 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
  83                         offset += copied;
  84                 } else {
  85                         pg++;
  86                         offset = 0;
  87                 }
  88         }
  89         return total_copied;
  90 }
  91
  92 /*
  93  * unlocks pages after btrfs_file_write is done with them
  94  */
  95 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
  96 {
  97         size_t i;
  98         for (i = 0; i < num_pages; i++) {
  99                 if (!pages[i])
 100                         break;
 101                 /* page checked is some magic around finding pages that
 102                  * have been modified without going through btrfs_set_page_dirty
 103                  * clear it here
 104                  */
 105                 ClearPageChecked(pages[i]);
 106                 unlock_page(pages[i]);
 107                 mark_page_accessed(pages[i]);
 108                 page_cache_release(pages[i]);
 109         }
 110 }
 111
 112 /*
 113  * after copy_from_user, pages need to be dirtied and we need to make
 114  * sure holes are created between the current EOF and the start of
 115  * any next extents (if required).
 116  *
 117  * this also makes the decision about creating an inline extent vs
 118  * doing real data extents, marking pages dirty and delalloc as required.
 119  */
 120 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 121                                    struct btrfs_root *root,
 122                                    struct file *file,
 123                                    struct page **pages,
 124                                    size_t num_pages,
 125                                    loff_t pos,
 126                                    size_t write_bytes)
 127 {
 128         int err = 0;
 129         int i;
 130         struct inode *inode = fdentry(file)->d_inode;
 131         u64 num_bytes;
 132         u64 start_pos;
 133         u64 end_of_last_block;
 134         u64 end_pos = pos + write_bytes;
 135         loff_t isize = i_size_read(inode);
 136
 137         start_pos = pos & ~((u64)root->sectorsize - 1);
 138         num_bytes = (write_bytes + pos - start_pos +
 139                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 140
 141         end_of_last_block = start_pos + num_bytes - 1;
 142         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 143                                         NULL);
 144         BUG_ON(err);
 145
 146         for (i = 0; i < num_pages; i++) {
 147                 struct page *p = pages[i];
 148                 SetPageUptodate(p);
 149                 ClearPageChecked(p);
 150                 set_page_dirty(p);
 151         }
 152         if (end_pos > isize) {
 153                 i_size_write(inode, end_pos);
 154                 /* we've only changed i_size in ram, and we haven't updated
 155                  * the disk i_size.  There is no need to log the inode
 156                  * at this time.
 157                  */
 158         }
 159         return 0;
 160 }
 161
 162 /*
 163  * this drops all the extents in the cache that intersect the range
 164  * [start, end].  Existing extents are split as required.
 165  */
 166 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 167                             int skip_pinned)
 168 {
 169         struct extent_map *em;
 170         struct extent_map *split = NULL;
 171         struct extent_map *split2 = NULL;
 172         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 173         u64 len = end - start + 1;
 174         int ret;
 175         int testend = 1;
 176         unsigned long flags;
 177         int compressed = 0;
 178
 179         WARN_ON(end < start);
 180         if (end == (u64)-1) {
 181                 len = (u64)-1;
 182                 testend = 0;
 183         }
 184         while (1) {
 185                 if (!split)
 186                         split = alloc_extent_map(GFP_NOFS);
 187                 if (!split2)
 188                         split2 = alloc_extent_map(GFP_NOFS);
 189
 190                 write_lock(&em_tree->lock);
 191                 em = lookup_extent_mapping(em_tree, start, len);
 192                 if (!em) {
 193                         write_unlock(&em_tree->lock);
 194                         break;
 195                 }
 196                 flags = em->flags;
 197                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 198                         if (testend && em->start + em->len >= start + len) {
 199                                 free_extent_map(em);
 200                                 write_unlock(&em_tree->lock);
 201                                 break;
 202                         }
 203                         start = em->start + em->len;
 204                         if (testend)
 205                                 len = start + len - (em->start + em->len);
 206                         free_extent_map(em);
 207                         write_unlock(&em_tree->lock);
 208                         continue;
 209                 }
 210                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 211                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 212                 remove_extent_mapping(em_tree, em);
 213
 214                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 215                     em->start < start) {
 216                         split->start = em->start;
 217                         split->len = start - em->start;
 218                         split->orig_start = em->orig_start;
 219                         split->block_start = em->block_start;
 220
 221                         if (compressed)
 222                                 split->block_len = em->block_len;
 223                         else
 224                                 split->block_len = split->len;
 225
 226                         split->bdev = em->bdev;
 227                         split->flags = flags;
 228                         ret = add_extent_mapping(em_tree, split);
 229                         BUG_ON(ret);
 230                         free_extent_map(split);
 231                         split = split2;
 232                         split2 = NULL;
 233                 }
 234                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 235                     testend && em->start + em->len > start + len) {
 236                         u64 diff = start + len - em->start;
 237
 238                         split->start = start + len;
 239                         split->len = em->start + em->len - (start + len);
 240                         split->bdev = em->bdev;
 241                         split->flags = flags;
 242
 243                         if (compressed) {
 244                                 split->block_len = em->block_len;
 245                                 split->block_start = em->block_start;
 246                                 split->orig_start = em->orig_start;
 247                         } else {
 248                                 split->block_len = split->len;
 249                                 split->block_start = em->block_start + diff;
 250                                 split->orig_start = split->start;
 251                         }
 252
 253                         ret = add_extent_mapping(em_tree, split);
 254                         BUG_ON(ret);
 255                         free_extent_map(split);
 256                         split = NULL;
 257                 }
 258                 write_unlock(&em_tree->lock);
 259
 260                 /* once for us */
 261                 free_extent_map(em);
 262                 /* once for the tree*/
 263                 free_extent_map(em);
 264         }
 265         if (split)
 266                 free_extent_map(split);
 267         if (split2)
 268                 free_extent_map(split2);
 269         return 0;
 270 }
 271
 272 /*
 273  * this is very complex, but the basic idea is to drop all extents
 274  * in the range start - end.  hint_block is filled in with a block number
 275  * that would be a good hint to the block allocator for this file.
 276  *
 277  * If an extent intersects the range but is not entirely inside the range
 278  * it is either truncated or split.  Anything entirely inside the range
 279  * is deleted from the tree.
 280  */
 281 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
 282                        u64 start, u64 end, u64 *hint_byte, int drop_cache)
 283 {
 284         struct btrfs_root *root = BTRFS_I(inode)->root;
 285         struct extent_buffer *leaf;
 286         struct btrfs_file_extent_item *fi;
 287         struct btrfs_path *path;
 288         struct btrfs_key key;
 289         struct btrfs_key new_key;
 290         u64 search_start = start;
 291         u64 disk_bytenr = 0;
 292         u64 num_bytes = 0;
 293         u64 extent_offset = 0;
 294         u64 extent_end = 0;
 295         int del_nr = 0;
 296         int del_slot = 0;
 297         int extent_type;
 298         int recow;
 299         int ret;
 300
 301         if (drop_cache)
 302                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 303
 304         path = btrfs_alloc_path();
 305         if (!path)
 306                 return -ENOMEM;
 307
 308         while (1) {
 309                 recow = 0;
 310                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 311                                                search_start, -1);
 312                 if (ret < 0)
 313                         break;
 314                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 315                         leaf = path->nodes[0];
 316                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 317                         if (key.objectid == inode->i_ino &&
 318                             key.type == BTRFS_EXTENT_DATA_KEY)
 319                                 path->slots[0]--;
 320                 }
 321                 ret = 0;
 322 next_slot:
 323                 leaf = path->nodes[0];
 324                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 325                         BUG_ON(del_nr > 0);
 326                         ret = btrfs_next_leaf(root, path);
 327                         if (ret < 0)
 328                                 break;
 329                         if (ret > 0) {
 330                                 ret = 0;
 331                                 break;
 332                         }
 333                         leaf = path->nodes[0];
 334                         recow = 1;
 335                 }
 336
 337                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 338                 if (key.objectid > inode->i_ino ||
 339                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 340                         break;
 341
 342                 fi = btrfs_item_ptr(leaf, path->slots[0],
 343                                     struct btrfs_file_extent_item);
 344                 extent_type = btrfs_file_extent_type(leaf, fi);
 345
 346                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 347                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 348                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 349                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 350                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 351                         extent_end = key.offset +
 352                                 btrfs_file_extent_num_bytes(leaf, fi);
 353                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 354                         extent_end = key.offset +
 355                                 btrfs_file_extent_inline_len(leaf, fi);
 356                 } else {
 357                         WARN_ON(1);
 358                         extent_end = search_start;
 359                 }
 360
 361                 if (extent_end <= search_start) {
 362                         path->slots[0]++;
 363                         goto next_slot;
 364                 }
 365
 366                 search_start = max(key.offset, start);
 367                 if (recow) {
 368                         btrfs_release_path(root, path);
 369                         continue;
 370                 }
 371
 372                 /*
 373                  *     | - range to drop - |
 374                  *  | -------- extent -------- |
 375                  */
 376                 if (start > key.offset && end < extent_end) {
 377                         BUG_ON(del_nr > 0);
 378                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 379
 380                         memcpy(&new_key, &key, sizeof(new_key));
 381                         new_key.offset = start;
 382                         ret = btrfs_duplicate_item(trans, root, path,
 383                                                    &new_key);
 384                         if (ret == -EAGAIN) {
 385                                 btrfs_release_path(root, path);
 386                                 continue;
 387                         }
 388                         if (ret < 0)
 389                                 break;
 390
 391                         leaf = path->nodes[0];
 392                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 393                                             struct btrfs_file_extent_item);
 394                         btrfs_set_file_extent_num_bytes(leaf, fi,
 395                                                         start - key.offset);
 396
 397                         fi = btrfs_item_ptr(leaf, path->slots[0],
 398                                             struct btrfs_file_extent_item);
 399
 400                         extent_offset += start - key.offset;
 401                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 402                         btrfs_set_file_extent_num_bytes(leaf, fi,
 403                                                         extent_end - start);
 404                         btrfs_mark_buffer_dirty(leaf);
 405
 406                         if (disk_bytenr > 0) {
 407                                 ret = btrfs_inc_extent_ref(trans, root,
 408                                                 disk_bytenr, num_bytes, 0,
 409                                                 root->root_key.objectid,
 410                                                 new_key.objectid,
 411                                                 start - extent_offset);
 412                                 BUG_ON(ret);
 413                                 *hint_byte = disk_bytenr;
 414                         }
 415                         key.offset = start;
 416                 }
 417                 /*
 418                  *  | ---- range to drop ----- |
 419                  *      | -------- extent -------- |
 420                  */
 421                 if (start <= key.offset && end < extent_end) {
 422                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 423
 424                         memcpy(&new_key, &key, sizeof(new_key));
 425                         new_key.offset = end;
 426                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 427
 428                         extent_offset += end - key.offset;
 429                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 430                         btrfs_set_file_extent_num_bytes(leaf, fi,
 431                                                         extent_end - end);
 432                         btrfs_mark_buffer_dirty(leaf);
 433                         if (disk_bytenr > 0) {
 434                                 inode_sub_bytes(inode, end - key.offset);
 435                                 *hint_byte = disk_bytenr;
 436                         }
 437                         break;
 438                 }
 439
 440                 search_start = extent_end;
 441                 /*
 442                  *       | ---- range to drop ----- |
 443                  *  | -------- extent -------- |
 444                  */
 445                 if (start > key.offset && end >= extent_end) {
 446                         BUG_ON(del_nr > 0);
 447                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 448
 449                         btrfs_set_file_extent_num_bytes(leaf, fi,
 450                                                         start - key.offset);
 451                         btrfs_mark_buffer_dirty(leaf);
 452                         if (disk_bytenr > 0) {
 453                                 inode_sub_bytes(inode, extent_end - start);
 454                                 *hint_byte = disk_bytenr;
 455                         }
 456                         if (end == extent_end)
 457                                 break;
 458
 459                         path->slots[0]++;
 460                         goto next_slot;
 461                 }
 462
 463                 /*
 464                  *  | ---- range to drop ----- |
 465                  *    | ------ extent ------ |
 466                  */
 467                 if (start <= key.offset && end >= extent_end) {
 468                         if (del_nr == 0) {
 469                                 del_slot = path->slots[0];
 470                                 del_nr = 1;
 471                         } else {
 472                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 473                                 del_nr++;
 474                         }
 475
 476                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 477                                 inode_sub_bytes(inode,
 478                                                 extent_end - key.offset);
 479                                 extent_end = ALIGN(extent_end,
 480                                                    root->sectorsize);
 481                         } else if (disk_bytenr > 0) {
 482                                 ret = btrfs_free_extent(trans, root,
 483                                                 disk_bytenr, num_bytes, 0,
 484                                                 root->root_key.objectid,
 485                                                 key.objectid, key.offset -
 486                                                 extent_offset);
 487                                 BUG_ON(ret);
 488                                 inode_sub_bytes(inode,
 489                                                 extent_end - key.offset);
 490                                 *hint_byte = disk_bytenr;
 491                         }
 492
 493                         if (end == extent_end)
 494                                 break;
 495
 496                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 497                                 path->slots[0]++;
 498                                 goto next_slot;
 499                         }
 500
 501                         ret = btrfs_del_items(trans, root, path, del_slot,
 502                                               del_nr);
 503                         BUG_ON(ret);
 504
 505                         del_nr = 0;
 506                         del_slot = 0;
 507
 508                         btrfs_release_path(root, path);
 509                         continue;
 510                 }
 511
 512                 BUG_ON(1);
 513         }
 514
 515         if (del_nr > 0) {
 516                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 517                 BUG_ON(ret);
 518         }
 519
 520         btrfs_free_path(path);
 521         return ret;
 522 }
 523
 524 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 525                             u64 objectid, u64 bytenr, u64 orig_offset,
 526                             u64 *start, u64 *end)
 527 {
 528         struct btrfs_file_extent_item *fi;
 529         struct btrfs_key key;
 530         u64 extent_end;
 531
 532         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 533                 return 0;
 534
 535         btrfs_item_key_to_cpu(leaf, &key, slot);
 536         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 537                 return 0;
 538
 539         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 540         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 541             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 542             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
 543             btrfs_file_extent_compression(leaf, fi) ||
 544             btrfs_file_extent_encryption(leaf, fi) ||
 545             btrfs_file_extent_other_encoding(leaf, fi))
 546                 return 0;
 547
 548         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 549         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 550                 return 0;
 551
 552         *start = key.offset;
 553         *end = extent_end;
 554         return 1;
 555 }
 556
 557 /*
 558  * Mark extent in the range start - end as written.
 559  *
 560  * This changes extent type from 'pre-allocated' to 'regular'. If only
 561  * part of extent is marked as written, the extent will be split into
 562  * two or three.
 563  */
 564 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 565                               struct inode *inode, u64 start, u64 end)
 566 {
 567         struct btrfs_root *root = BTRFS_I(inode)->root;
 568         struct extent_buffer *leaf;
 569         struct btrfs_path *path;
 570         struct btrfs_file_extent_item *fi;
 571         struct btrfs_key key;
 572         struct btrfs_key new_key;
 573         u64 bytenr;
 574         u64 num_bytes;
 575         u64 extent_end;
 576         u64 orig_offset;
 577         u64 other_start;
 578         u64 other_end;
 579         u64 split;
 580         int del_nr = 0;
 581         int del_slot = 0;
 582         int recow;
 583         int ret;
 584
 585         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 586
 587         path = btrfs_alloc_path();
 588         BUG_ON(!path);
 589 again:
 590         recow = 0;
 591         split = start;
 592         key.objectid = inode->i_ino;
 593         key.type = BTRFS_EXTENT_DATA_KEY;
 594         key.offset = split;
 595
 596         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 597         if (ret > 0 && path->slots[0] > 0)
 598                 path->slots[0]--;
 599
 600         leaf = path->nodes[0];
 601         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 602         BUG_ON(key.objectid != inode->i_ino ||
 603                key.type != BTRFS_EXTENT_DATA_KEY);
 604         fi = btrfs_item_ptr(leaf, path->slots[0],
 605                             struct btrfs_file_extent_item);
 606         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
 607                BTRFS_FILE_EXTENT_PREALLOC);
 608         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 609         BUG_ON(key.offset > start || extent_end < end);
 610
 611         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 612         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 613         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 614         memcpy(&new_key, &key, sizeof(new_key));
 615
 616         if (start == key.offset && end < extent_end) {
 617                 other_start = 0;
 618                 other_end = start;
 619                 if (extent_mergeable(leaf, path->slots[0] - 1,
 620                                      inode->i_ino, bytenr, orig_offset,
 621                                      &other_start, &other_end)) {
 622                         new_key.offset = end;
 623                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 624                         fi = btrfs_item_ptr(leaf, path->slots[0],
 625                                             struct btrfs_file_extent_item);
 626                         btrfs_set_file_extent_num_bytes(leaf, fi,
 627                                                         extent_end - end);
 628                         btrfs_set_file_extent_offset(leaf, fi,
 629                                                      end - orig_offset);
 630                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 631                                             struct btrfs_file_extent_item);
 632                         btrfs_set_file_extent_num_bytes(leaf, fi,
 633                                                         end - other_start);
 634                         btrfs_mark_buffer_dirty(leaf);
 635                         goto out;
 636                 }
 637         }
 638
 639         if (start > key.offset && end == extent_end) {
 640                 other_start = end;
 641                 other_end = 0;
 642                 if (extent_mergeable(leaf, path->slots[0] + 1,
 643                                      inode->i_ino, bytenr, orig_offset,
 644                                      &other_start, &other_end)) {
 645                         fi = btrfs_item_ptr(leaf, path->slots[0],
 646                                             struct btrfs_file_extent_item);
 647                         btrfs_set_file_extent_num_bytes(leaf, fi,
 648                                                         start - key.offset);
 649                         path->slots[0]++;
 650                         new_key.offset = start;
 651                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 652
 653                         fi = btrfs_item_ptr(leaf, path->slots[0],
 654                                             struct btrfs_file_extent_item);
 655                         btrfs_set_file_extent_num_bytes(leaf, fi,
 656                                                         other_end - start);
 657                         btrfs_set_file_extent_offset(leaf, fi,
 658                                                      start - orig_offset);
 659                         btrfs_mark_buffer_dirty(leaf);
 660                         goto out;
 661                 }
 662         }
 663
 664         while (start > key.offset || end < extent_end) {
 665                 if (key.offset == start)
 666                         split = end;
 667
 668                 new_key.offset = split;
 669                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
 670                 if (ret == -EAGAIN) {
 671                         btrfs_release_path(root, path);
 672                         goto again;
 673                 }
 674                 BUG_ON(ret < 0);
 675
 676                 leaf = path->nodes[0];
 677                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 678                                     struct btrfs_file_extent_item);
 679                 btrfs_set_file_extent_num_bytes(leaf, fi,
 680                                                 split - key.offset);
 681
 682                 fi = btrfs_item_ptr(leaf, path->slots[0],
 683                                     struct btrfs_file_extent_item);
 684
 685                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
 686                 btrfs_set_file_extent_num_bytes(leaf, fi,
 687                                                 extent_end - split);
 688                 btrfs_mark_buffer_dirty(leaf);
 689
 690                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 691                                            root->root_key.objectid,
 692                                            inode->i_ino, orig_offset);
 693                 BUG_ON(ret);
 694
 695                 if (split == start) {
 696                         key.offset = start;
 697                 } else {
 698                         BUG_ON(start != key.offset);
 699                         path->slots[0]--;
 700                         extent_end = end;
 701                 }
 702                 recow = 1;
 703         }
 704
 705         other_start = end;
 706         other_end = 0;
 707         if (extent_mergeable(leaf, path->slots[0] + 1,
 708                              inode->i_ino, bytenr, orig_offset,
 709                              &other_start, &other_end)) {
 710                 if (recow) {
 711                         btrfs_release_path(root, path);
 712                         goto again;
 713                 }
 714                 extent_end = other_end;
 715                 del_slot = path->slots[0] + 1;
 716                 del_nr++;
 717                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 718                                         0, root->root_key.objectid,
 719                                         inode->i_ino, orig_offset);
 720                 BUG_ON(ret);
 721         }
 722         other_start = 0;
 723         other_end = start;
 724         if (extent_mergeable(leaf, path->slots[0] - 1,
 725                              inode->i_ino, bytenr, orig_offset,
 726                              &other_start, &other_end)) {
 727                 if (recow) {
 728                         btrfs_release_path(root, path);
 729                         goto again;
 730                 }
 731                 key.offset = other_start;
 732                 del_slot = path->slots[0];
 733                 del_nr++;
 734                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 735                                         0, root->root_key.objectid,
 736                                         inode->i_ino, orig_offset);
 737                 BUG_ON(ret);
 738         }
 739         if (del_nr == 0) {
 740                 fi = btrfs_item_ptr(leaf, path->slots[0],
 741                            struct btrfs_file_extent_item);
 742                 btrfs_set_file_extent_type(leaf, fi,
 743                                            BTRFS_FILE_EXTENT_REG);
 744                 btrfs_mark_buffer_dirty(leaf);
 745         } else {
 746                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 747                            struct btrfs_file_extent_item);
 748                 btrfs_set_file_extent_type(leaf, fi,
 749                                            BTRFS_FILE_EXTENT_REG);
 750                 btrfs_set_file_extent_num_bytes(leaf, fi,
 751                                                 extent_end - key.offset);
 752                 btrfs_mark_buffer_dirty(leaf);
 753
 754                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 755                 BUG_ON(ret);
 756         }
 757 out:
 758         btrfs_free_path(path);
 759         return 0;
 760 }
 761
 762 /*
 763  * this gets pages into the page cache and locks them down, it also properly
 764  * waits for data=ordered extents to finish before allowing the pages to be
 765  * modified.
 766  */
 767 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 768                          struct page **pages, size_t num_pages,
 769                          loff_t pos, unsigned long first_index,
 770                          unsigned long last_index, size_t write_bytes)
 771 {
 772         struct extent_state *cached_state = NULL;
 773         int i;
 774         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 775         struct inode *inode = fdentry(file)->d_inode;
 776         int err = 0;
 777         u64 start_pos;
 778         u64 last_pos;
 779
 780         start_pos = pos & ~((u64)root->sectorsize - 1);
 781         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 782
 783         if (start_pos > inode->i_size) {
 784                 err = btrfs_cont_expand(inode, start_pos);
 785                 if (err)
 786                         return err;
 787         }
 788
 789         memset(pages, 0, num_pages * sizeof(struct page *));
 790 again:
 791         for (i = 0; i < num_pages; i++) {
 792                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 793                 if (!pages[i]) {
 794                         err = -ENOMEM;
 795                         BUG_ON(1);
 796                 }
 797                 wait_on_page_writeback(pages[i]);
 798         }
 799         if (start_pos < inode->i_size) {
 800                 struct btrfs_ordered_extent *ordered;
 801                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
 802                                  start_pos, last_pos - 1, 0, &cached_state,
 803                                  GFP_NOFS);
 804                 ordered = btrfs_lookup_first_ordered_extent(inode,
 805                                                             last_pos - 1);
 806                 if (ordered &&
 807                     ordered->file_offset + ordered->len > start_pos &&
 808                     ordered->file_offset < last_pos) {
 809                         btrfs_put_ordered_extent(ordered);
 810                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 811                                              start_pos, last_pos - 1,
 812                                              &cached_state, GFP_NOFS);
 813                         for (i = 0; i < num_pages; i++) {
 814                                 unlock_page(pages[i]);
 815                                 page_cache_release(pages[i]);
 816                         }
 817                         btrfs_wait_ordered_range(inode, start_pos,
 818                                                  last_pos - start_pos);
 819                         goto again;
 820                 }
 821                 if (ordered)
 822                         btrfs_put_ordered_extent(ordered);
 823
 824                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
 825                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 826                                   EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
 827                                   GFP_NOFS);
 828                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 829                                      start_pos, last_pos - 1, &cached_state,
 830                                      GFP_NOFS);
 831         }
 832         for (i = 0; i < num_pages; i++) {
 833                 clear_page_dirty_for_io(pages[i]);
 834                 set_page_extent_mapped(pages[i]);
 835                 WARN_ON(!PageLocked(pages[i]));
 836         }
 837         return 0;
 838 }
 839
 840 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
 841                                     const struct iovec *iov,
 842                                     unsigned long nr_segs, loff_t pos)
 843 {
 844         struct file *file = iocb->ki_filp;
 845         struct inode *inode = fdentry(file)->d_inode;
 846         struct btrfs_root *root = BTRFS_I(inode)->root;
 847         struct page *pinned[2];
 848         struct page **pages = NULL;
 849         struct iov_iter i;
 850         loff_t *ppos = &iocb->ki_pos;
 851         loff_t start_pos;
 852         ssize_t num_written = 0;
 853         ssize_t err = 0;
 854         size_t count;
 855         size_t ocount;
 856         int ret = 0;
 857         int nrptrs;
 858         unsigned long first_index;
 859         unsigned long last_index;
 860         int will_write;
 861         int buffered = 0;
 862         int copied = 0;
 863         int dirty_pages = 0;
 864
 865         will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
 866                       (file->f_flags & O_DIRECT));
 867
 868         pinned[0] = NULL;
 869         pinned[1] = NULL;
 870
 871         start_pos = pos;
 872
 873         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 874
 875         mutex_lock(&inode->i_mutex);
 876
 877         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
 878         if (err)
 879                 goto out;
 880         count = ocount;
 881
 882         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 883         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 884         if (err)
 885                 goto out;
 886
 887         if (count == 0)
 888                 goto out;
 889
 890         err = file_remove_suid(file);
 891         if (err)
 892                 goto out;
 893
 894         file_update_time(file);
 895         BTRFS_I(inode)->sequence++;
 896
 897         if (unlikely(file->f_flags & O_DIRECT)) {
 898                 num_written = generic_file_direct_write(iocb, iov, &nr_segs,
 899                                                         pos, ppos, count,
 900                                                         ocount);
 901                 /*
 902                  * the generic O_DIRECT will update in-memory i_size after the
 903                  * DIOs are done.  But our endio handlers that update the on
 904                  * disk i_size never update past the in memory i_size.  So we
 905                  * need one more update here to catch any additions to the
 906                  * file
 907                  */
 908                 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
 909                         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
 910                         mark_inode_dirty(inode);
 911                 }
 912
 913                 if (num_written < 0) {
 914                         ret = num_written;
 915                         num_written = 0;
 916                         goto out;
 917                 } else if (num_written == count) {
 918                         /* pick up pos changes done by the generic code */
 919                         pos = *ppos;
 920                         goto out;
 921                 }
 922                 /*
 923                  * We are going to do buffered for the rest of the range, so we
 924                  * need to make sure to invalidate the buffered pages when we're
 925                  * done.
 926                  */
 927                 buffered = 1;
 928                 pos += num_written;
 929         }
 930
 931         iov_iter_init(&i, iov, nr_segs, count, num_written);
 932         nrptrs = min((iov_iter_count(&i) + PAGE_CACHE_SIZE - 1) /
 933                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
 934                      (sizeof(struct page *)));
 935         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 936
 937         /* generic_write_checks can change our pos */
 938         start_pos = pos;
 939
 940         first_index = pos >> PAGE_CACHE_SHIFT;
 941         last_index = (pos + iov_iter_count(&i)) >> PAGE_CACHE_SHIFT;
 942
 943         /*
 944          * there are lots of better ways to do this, but this code
 945          * makes sure the first and last page in the file range are
 946          * up to date and ready for cow
 947          */
 948         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 949                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 950                 if (!PageUptodate(pinned[0])) {
 951                         ret = btrfs_readpage(NULL, pinned[0]);
 952                         BUG_ON(ret);
 953                         wait_on_page_locked(pinned[0]);
 954                 } else {
 955                         unlock_page(pinned[0]);
 956                 }
 957         }
 958         if ((pos + iov_iter_count(&i)) & (PAGE_CACHE_SIZE - 1)) {
 959                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 960                 if (!PageUptodate(pinned[1])) {
 961                         ret = btrfs_readpage(NULL, pinned[1]);
 962                         BUG_ON(ret);
 963                         wait_on_page_locked(pinned[1]);
 964                 } else {
 965                         unlock_page(pinned[1]);
 966                 }
 967         }
 968
 969         while (iov_iter_count(&i) > 0) {
 970                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 971                 size_t write_bytes = min(iov_iter_count(&i),
 972                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
 973                                          offset);
 974                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 975                                         PAGE_CACHE_SHIFT;
 976
 977                 WARN_ON(num_pages > nrptrs);
 978                 memset(pages, 0, sizeof(struct page *) * nrptrs);
 979
 980                 /*
 981                  * Fault pages before locking them in prepare_pages
 982                  * to avoid recursive lock
 983                  */
 984                 if (unlikely(iov_iter_fault_in_readable(&i, write_bytes))) {
 985                         ret = -EFAULT;
 986                         goto out;
 987                 }
 988
 989                 ret = btrfs_delalloc_reserve_space(inode,
 990                                         num_pages << PAGE_CACHE_SHIFT);
 991                 if (ret)
 992                         goto out;
 993
 994                 ret = prepare_pages(root, file, pages, num_pages,
 995                                     pos, first_index, last_index,
 996                                     write_bytes);
 997                 if (ret) {
 998                         btrfs_delalloc_release_space(inode,
 999                                         num_pages << PAGE_CACHE_SHIFT);
1000                         goto out;
1001                 }
1002
1003                 copied = btrfs_copy_from_user(pos, num_pages,
1004                                            write_bytes, pages, &i);
1005                 dirty_pages = (copied + PAGE_CACHE_SIZE - 1) >>
1006                                         PAGE_CACHE_SHIFT;
1007
1008                 if (num_pages > dirty_pages) {
1009                         if (copied > 0)
1010                                 atomic_inc(
1011                                         &BTRFS_I(inode)->outstanding_extents);
1012                         btrfs_delalloc_release_space(inode,
1013                                         (num_pages - dirty_pages) <<
1014                                         PAGE_CACHE_SHIFT);
1015                 }
1016
1017                 if (copied > 0) {
1018                         dirty_and_release_pages(NULL, root, file, pages,
1019                                                 dirty_pages, pos, copied);
1020                 }
1021
1022                 btrfs_drop_pages(pages, num_pages);
1023
1024                 if (copied > 0) {
1025                         if (will_write) {
1026                                 filemap_fdatawrite_range(inode->i_mapping, pos,
1027                                                          pos + copied - 1);
1028                         } else {
1029                                 balance_dirty_pages_ratelimited_nr(
1030                                                         inode->i_mapping,
1031                                                         dirty_pages);
1032                                 if (dirty_pages <
1033                                 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1034                                         btrfs_btree_balance_dirty(root, 1);
1035                                 btrfs_throttle(root);
1036                         }
1037                 }
1038
1039                 pos += copied;
1040                 num_written += copied;
1041
1042                 cond_resched();
1043         }
1044 out:
1045         mutex_unlock(&inode->i_mutex);
1046         if (ret)
1047                 err = ret;
1048
1049         kfree(pages);
1050         if (pinned[0])
1051                 page_cache_release(pinned[0]);
1052         if (pinned[1])
1053                 page_cache_release(pinned[1]);
1054         *ppos = pos;
1055
1056         /*
1057          * we want to make sure fsync finds this change
1058          * but we haven't joined a transaction running right now.
1059          *
1060          * Later on, someone is sure to update the inode and get the
1061          * real transid recorded.
1062          *
1063          * We set last_trans now to the fs_info generation + 1,
1064          * this will either be one more than the running transaction
1065          * or the generation used for the next transaction if there isn't
1066          * one running right now.
1067          */
1068         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1069
1070         if (num_written > 0 && will_write) {
1071                 struct btrfs_trans_handle *trans;
1072
1073                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1074                 if (err)
1075                         num_written = err;
1076
1077                 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1078                         trans = btrfs_start_transaction(root, 0);
1079                         if (IS_ERR(trans)) {
1080                                 num_written = PTR_ERR(trans);
1081                                 goto done;
1082                         }
1083                         mutex_lock(&inode->i_mutex);
1084                         ret = btrfs_log_dentry_safe(trans, root,
1085                                                     file->f_dentry);
1086                         mutex_unlock(&inode->i_mutex);
1087                         if (ret == 0) {
1088                                 ret = btrfs_sync_log(trans, root);
1089                                 if (ret == 0)
1090                                         btrfs_end_transaction(trans, root);
1091                                 else
1092                                         btrfs_commit_transaction(trans, root);
1093                         } else if (ret != BTRFS_NO_LOG_SYNC) {
1094                                 btrfs_commit_transaction(trans, root);
1095                         } else {
1096                                 btrfs_end_transaction(trans, root);
1097                         }
1098                 }
1099                 if (file->f_flags & O_DIRECT && buffered) {
1100                         invalidate_mapping_pages(inode->i_mapping,
1101                               start_pos >> PAGE_CACHE_SHIFT,
1102                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1103                 }
1104         }
1105 done:
1106         current->backing_dev_info = NULL;
1107         return num_written ? num_written : err;
1108 }
1109
1110 int btrfs_release_file(struct inode *inode, struct file *filp)
1111 {
1112         /*
1113          * ordered_data_close is set by settattr when we are about to truncate
1114          * a file from a non-zero size to a zero size.  This tries to
1115          * flush down new bytes that may have been written if the
1116          * application were using truncate to replace a file in place.
1117          */
1118         if (BTRFS_I(inode)->ordered_data_close) {
1119                 BTRFS_I(inode)->ordered_data_close = 0;
1120                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1121                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1122                         filemap_flush(inode->i_mapping);
1123         }
1124         if (filp->private_data)
1125                 btrfs_ioctl_trans_end(filp);
1126         return 0;
1127 }
1128
1129 /*
1130  * fsync call for both files and directories.  This logs the inode into
1131  * the tree log instead of forcing full commits whenever possible.
1132  *
1133  * It needs to call filemap_fdatawait so that all ordered extent updates are
1134  * in the metadata btree are up to date for copying to the log.
1135  *
1136  * It drops the inode mutex before doing the tree log commit.  This is an
1137  * important optimization for directories because holding the mutex prevents
1138  * new operations on the dir while we write to disk.
1139  */
1140 int btrfs_sync_file(struct file *file, int datasync)
1141 {
1142         struct dentry *dentry = file->f_path.dentry;
1143         struct inode *inode = dentry->d_inode;
1144         struct btrfs_root *root = BTRFS_I(inode)->root;
1145         int ret = 0;
1146         struct btrfs_trans_handle *trans;
1147
1148
1149         /* we wait first, since the writeback may change the inode */
1150         root->log_batch++;
1151         /* the VFS called filemap_fdatawrite for us */
1152         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1153         root->log_batch++;
1154
1155         /*
1156          * check the transaction that last modified this inode
1157          * and see if its already been committed
1158          */
1159         if (!BTRFS_I(inode)->last_trans)
1160                 goto out;
1161
1162         /*
1163          * if the last transaction that changed this file was before
1164          * the current transaction, we can bail out now without any
1165          * syncing
1166          */
1167         mutex_lock(&root->fs_info->trans_mutex);
1168         if (BTRFS_I(inode)->last_trans <=
1169             root->fs_info->last_trans_committed) {
1170                 BTRFS_I(inode)->last_trans = 0;
1171                 mutex_unlock(&root->fs_info->trans_mutex);
1172                 goto out;
1173         }
1174         mutex_unlock(&root->fs_info->trans_mutex);
1175
1176         /*
1177          * ok we haven't committed the transaction yet, lets do a commit
1178          */
1179         if (file->private_data)
1180                 btrfs_ioctl_trans_end(file);
1181
1182         trans = btrfs_start_transaction(root, 0);
1183         if (IS_ERR(trans)) {
1184                 ret = PTR_ERR(trans);
1185                 goto out;
1186         }
1187
1188         ret = btrfs_log_dentry_safe(trans, root, dentry);
1189         if (ret < 0)
1190                 goto out;
1191
1192         /* we've logged all the items and now have a consistent
1193          * version of the file in the log.  It is possible that
1194          * someone will come in and modify the file, but that's
1195          * fine because the log is consistent on disk, and we
1196          * have references to all of the file's extents
1197          *
1198          * It is possible that someone will come in and log the
1199          * file again, but that will end up using the synchronization
1200          * inside btrfs_sync_log to keep things safe.
1201          */
1202         mutex_unlock(&dentry->d_inode->i_mutex);
1203
1204         if (ret != BTRFS_NO_LOG_SYNC) {
1205                 if (ret > 0) {
1206                         ret = btrfs_commit_transaction(trans, root);
1207                 } else {
1208                         ret = btrfs_sync_log(trans, root);
1209                         if (ret == 0)
1210                                 ret = btrfs_end_transaction(trans, root);
1211                         else
1212                                 ret = btrfs_commit_transaction(trans, root);
1213                 }
1214         } else {
1215                 ret = btrfs_end_transaction(trans, root);
1216         }
1217         mutex_lock(&dentry->d_inode->i_mutex);
1218 out:
1219         return ret > 0 ? -EIO : ret;
1220 }
1221
1222 static const struct vm_operations_struct btrfs_file_vm_ops = {
1223         .fault          = filemap_fault,
1224         .page_mkwrite   = btrfs_page_mkwrite,
1225 };
1226
1227 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1228 {
1229         struct address_space *mapping = filp->f_mapping;
1230
1231         if (!mapping->a_ops->readpage)
1232                 return -ENOEXEC;
1233
1234         file_accessed(filp);
1235         vma->vm_ops = &btrfs_file_vm_ops;
1236         vma->vm_flags |= VM_CAN_NONLINEAR;
1237
1238         return 0;
1239 }
1240
1241 static long btrfs_fallocate(struct file *file, int mode,
1242                             loff_t offset, loff_t len)
1243 {
1244         struct inode *inode = file->f_path.dentry->d_inode;
1245         struct extent_state *cached_state = NULL;
1246         u64 cur_offset;
1247         u64 last_byte;
1248         u64 alloc_start;
1249         u64 alloc_end;
1250         u64 alloc_hint = 0;
1251         u64 locked_end;
1252         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1253         struct extent_map *em;
1254         int ret;
1255
1256         alloc_start = offset & ~mask;
1257         alloc_end =  (offset + len + mask) & ~mask;
1258
1259         /* We only support the FALLOC_FL_KEEP_SIZE mode */
1260         if (mode & ~FALLOC_FL_KEEP_SIZE)
1261                 return -EOPNOTSUPP;
1262
1263         /*
1264          * wait for ordered IO before we have any locks.  We'll loop again
1265          * below with the locks held.
1266          */
1267         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1268
1269         mutex_lock(&inode->i_mutex);
1270         ret = inode_newsize_ok(inode, alloc_end);
1271         if (ret)
1272                 goto out;
1273
1274         if (alloc_start > inode->i_size) {
1275                 ret = btrfs_cont_expand(inode, alloc_start);
1276                 if (ret)
1277                         goto out;
1278         }
1279
1280         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1281         if (ret)
1282                 goto out;
1283
1284         locked_end = alloc_end - 1;
1285         while (1) {
1286                 struct btrfs_ordered_extent *ordered;
1287
1288                 /* the extent lock is ordered inside the running
1289                  * transaction
1290                  */
1291                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1292                                  locked_end, 0, &cached_state, GFP_NOFS);
1293                 ordered = btrfs_lookup_first_ordered_extent(inode,
1294                                                             alloc_end - 1);
1295                 if (ordered &&
1296                     ordered->file_offset + ordered->len > alloc_start &&
1297                     ordered->file_offset < alloc_end) {
1298                         btrfs_put_ordered_extent(ordered);
1299                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1300                                              alloc_start, locked_end,
1301                                              &cached_state, GFP_NOFS);
1302                         /*
1303                          * we can't wait on the range with the transaction
1304                          * running or with the extent lock held
1305                          */
1306                         btrfs_wait_ordered_range(inode, alloc_start,
1307                                                  alloc_end - alloc_start);
1308                 } else {
1309                         if (ordered)
1310                                 btrfs_put_ordered_extent(ordered);
1311                         break;
1312                 }
1313         }
1314
1315         cur_offset = alloc_start;
1316         while (1) {
1317                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1318                                       alloc_end - cur_offset, 0);
1319                 BUG_ON(IS_ERR(em) || !em);
1320                 last_byte = min(extent_map_end(em), alloc_end);
1321                 last_byte = (last_byte + mask) & ~mask;
1322                 if (em->block_start == EXTENT_MAP_HOLE ||
1323                     (cur_offset >= inode->i_size &&
1324                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1325                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1326                                                         last_byte - cur_offset,
1327                                                         1 << inode->i_blkbits,
1328                                                         offset + len,
1329                                                         &alloc_hint);
1330                         if (ret < 0) {
1331                                 free_extent_map(em);
1332                                 break;
1333                         }
1334                 }
1335                 free_extent_map(em);
1336
1337                 cur_offset = last_byte;
1338                 if (cur_offset >= alloc_end) {
1339                         ret = 0;
1340                         break;
1341                 }
1342         }
1343         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1344                              &cached_state, GFP_NOFS);
1345
1346         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1347 out:
1348         mutex_unlock(&inode->i_mutex);
1349         return ret;
1350 }
1351
1352 const struct file_operations btrfs_file_operations = {
1353         .llseek         = generic_file_llseek,
1354         .read           = do_sync_read,
1355         .write          = do_sync_write,
1356         .aio_read       = generic_file_aio_read,
1357         .splice_read    = generic_file_splice_read,
1358         .aio_write      = btrfs_file_aio_write,
1359         .mmap           = btrfs_file_mmap,
1360         .open           = generic_file_open,
1361         .release        = btrfs_release_file,
1362         .fsync          = btrfs_sync_file,
1363         .fallocate      = btrfs_fallocate,
1364         .unlocked_ioctl = btrfs_ioctl,
1365 #ifdef CONFIG_COMPAT
1366         .compat_ioctl   = btrfs_ioctl,
1367 #endif
1368 };